Recent years have seen a growing array of activities in developing statistical models for modelling real-life networks. Since many of these networks are sparse, an all too often practice in the literature is to apply a developed model to a subnetwork typically by discarding nodes due to their lack of connectivity. In this note, we provide the first result highlighting issues with this practice which we call the fallacy of data-selective inference. We demonstrate this fallacy by examining the estimation bias in the Erdős–Rényi model theoretically and in the stochastic block model empirically.

中文翻译：

---

建模网络中数据选择性推理的谬误

近年来，在开发用于模拟现实网络的统计模型方面的活动越来越多。由于这些网络中的许多网络都是稀疏的，因此文献中经常采用的做法是将开发的模型应用于子网络，通常是通过丢弃由于缺乏连接性而导致的节点。在这篇文章中，我们提供了第一个结果，强调了这种做法的问题，我们称之为数据选择性推理的谬误。我们通过从理论上检查 Erdős–Rényi 模型中的估计偏差以及从经验上检查随机块模型中的估计偏差来证明这一谬误。